Data-Driven Methods for Robust Estimation of Site-Specific Ground Motions

Abstract

Accurately estimating site effects is critical for predicting earthquake ground motions for site- specific applications. The current state-of-the-practice for obtaining design ground motions involve estimating site effects using simple empirical models based on the time-averaged shear-wave velocity in the top 30-m (VS30) or performing site response analyses, both have certain drawbacks that lead to large uncertainties in their predictions. Furthermore, the existing approaches require dynamic site characterization that is often expensive and require skilled labor. When applied to distributed infrastructure, such as pipelines or major roads, the current methods of predicting site effects become not feasible for the level of accuracy that is often required. This presents a need for alternative approaches for the prediction of site effects that reduce the associated uncertainties while being cheaper and easier to implement in practice. The overall objective of the research presented in this dissertation is to develop new data-driven methods for easier and more accurate site response predictions. The research presents three new data-driven methods for the prediction of site effects. The first method involves inverting weak ground motions from small magnitude earthquakes to constrain site effects. With the abundance of small magnitude events in seismically active regions, such as California, the proposed approach utilizes the resulting weak ground motions for extracting site effects with temporary installment of seismic instruments. The second method uses microtremor horizonal-to-vertical spectral ratios (mHVSR) as an alternative site proxy for predicting site effects. To this effect, a curated database of mHVSR for permanent seismic stations in the United States has been compiled, followed by the development of new Artificial Neural Network (ANN) models using the compiled database for the prediction of site effects. It has been shown that mHVSR has similar predictive power compared to VS30. The third method focuses on the prediction of non-linear site effects. Based on 1D site response simulations, new models have been developed for predicting non-linear site terms. The models developed herein capture high frequency behavior more appropriately. Overall, the new data-driven methods presented in the dissertation lead to robust estimation of site effects.